On-Device AI (Apple Intelligence, Gemini Nano) for E-commerce Brands

Apple Intelligence is Apple’s set of AI features for recent iPhone, iPad and Mac models. Many tasks run on the device’s own chip, and heavier requests can go to Apple’s Private Cloud Compute servers. Features include notification and email summaries, writing tools, visual intelligence that identifies objects through the camera, and Siri improvements. Apple has also opened its on-device foundation model to app developers through a framework.

Gemini Nano is Google’s smallest Gemini model, built to run on Android phones such as recent Pixel and Samsung Galaxy devices. It powers features like message summaries, smart replies and scam detection, and developers can access it through Android’s AICore system service. Google has also tested built-in AI APIs in Chrome that run a Nano model locally for tasks such as summarising and translating.

Device support, languages and feature lists vary by region and change with each software release, so check current details before planning around a specific feature.

These features sit between your brand and your customer. A push notification about an order might be shown as a one-line summary next to other app alerts. A promotional email might be condensed before the customer opens it. A product seen in a shop window might be identified through the camera and searched directly.

You do not control these summaries, but you control the input. Messages and pages that are clear and specific are summarised accurately. Vague or clever copy can be summarised into something misleading.

Notifications and transactional messages

Put the key fact first: order shipped, delivery date, action needed. Avoid stacking several messages in a row that only make sense together, since summaries may merge them. Test how your app’s notifications look when grouped on a device with summaries switched on.

Email campaigns

Mail apps on iOS and Android can show summaries or priority labels. Write a subject line and first sentence that state the offer and the deadline plainly. Hidden preheader tricks and image-only emails give the device little real text to work with.

Visual search

Camera-based identification relies partly on web search in the background. Product images with clean backgrounds, consistent naming and complete product schema make it easier for search systems to connect a photo with your listing.

In-app features

Brands with their own apps can use on-device models for tasks such as summarising reviews, suggesting search terms or helping customers write a return reason. Because processing stays on the device, these features work offline and avoid sending personal text to a server.

• Small models are less capable than cloud models; they handle short summaries and rewrites, not complex reasoning.
• Only recent, higher-end devices support the features, so many customers will not see them.
• Language support is narrower than cloud services, and some features launch in English first.
• Users can switch features off, and app developers cannot force summaries on or off for their notifications.
• Model behaviour can change with an operating system update, without notice to brands.

• Audit the last thirty push notifications and rewrite any that hide the main fact.
• Check that every marketing email has a real text version and a clear first sentence.
• Review product titles for consistency: brand, product type, key attribute, size or colour.
• Complete Product schema with GTIN, brand and availability on all live products.
• Make sure pages are accessible, with proper headings and alt text; device features read the same structure screen readers use.
• If you have an app, list one task where an on-device model could help customers without sending data off the phone.

Visual lookups and assistant features change the path to a product page. A shopper might photograph a lamp in a friend’s flat, get a set of matching results and tap through without typing a search. The store that wins that click has clear product images, a descriptive title and structured data that ties the image to the item.

Assistants on the device can also act across apps, for example finding an order confirmation in email or adding a delivery date to the calendar. Well-structured transactional emails, using schema such as order markup where supported, make these actions more reliable for customers. Messy emails simply get skipped.

None of this replaces search engine optimisation. It adds another reader, a small model on the customer’s device, that rewards the same clarity search engines already prefer.